Atlas based breast registration and segmentation in the Mediolateral Oblique and Craniocaudal Views

In the clinical evaluation of the risk of breast cancer, breast density obtained from analyzing mammograms is of high importance. Higher breast density increases the likelihood of getting breast cancer and also the likelihood of radiologists missing small lesions. Consequently, we analyze breast mammograms to determine which mammograms need greater radiological review. In this paper, we come up with a new atlas-based segmentation technique to register and segment the breast region in mammograms. Atlas-based segmentation is a technique that seeks to segment the pertinent anatomy in medical images. This method is suitable for segmentation of images, whenever there are ambiguous relationships between regions and pixel intensities. This paper evaluates how an atlas-based method can be used to segment the breast region and breast tissue from a mammogram. In our dataset, we have exactly 50 CC images and 15 MLO images. We show the segmented images and, using the dice similarity coefficient to find the performance of the segmentation, show that the accuracy of the segmentation using atlas-based techniques are at 0.87.